From 1919eeccdd31e16ec7a44478b756bcd974c35a3c Mon Sep 17 00:00:00 2001 From: Lin Min Htoo Date: Wed, 17 Aug 2022 10:37:53 +0800 Subject: [PATCH] doc: add neuralsym repo link to README --- README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/README.md b/README.md index 19b996b..ffc0e54 100644 --- a/README.md +++ b/README.md @@ -190,6 +190,7 @@ which will output 3 cleaned CSV files in ``` rxnebm/data/cleaned_data ``` of the - To train NeuralSym, we simply use the 3 .pickle files ``` 50k_clean_rxnsmi_noreagent_allmapped_canon_.pickle ``` generated using ```clean_smiles.py``` above, which contain the extra-cleaned USPTO-50K reactions. They've also been placed in NeuralSym's input data folder ``` rxnebm/proposer/neuralsym/data/ ```. - As the original authors did not open-source NeuralSym, we re-implemented it from scratch following their paper and the repo is placed at:
``` cd rxnebm/proposer/neuralsym ```
+For reference, our re-implementation can also be found at: https://github.com/linminhtoo/neuralsym - To setup the environment for NeuralSym: once you're in ``` neuralsym ```, run:
``` bash setup.sh ```
This creates a conda environment called ```neuralsym```, which you need to activate to train/test/propose with NeuralSym. - To preprocess training data into 32681-dim fingerprints. As we've heavily optimized this step, it takes only ~10 mins on 16 cores, and probably ~15-20 mins on 8 cores.